To prevent an over-output of logs from affecting disk and system performance, Milvus by default outputs logs at the
info level while running. However, sometimes logs at the
info level are not sufficient enough to help us efficiently identify bugs and issues. What's worse, in some cases, changing the log level and restarting the service might lead to the failure of reproducing the issues, making troubleshooting all the more difficult. Consequently, the support for changing log levels dynamically in the Milvus vector database is urgently needed.
This article aims to introduce the mechanism behind that enables changing log levels dynamically and provide instructions on how to do so in the Milvus vector database.
The Milvus vector database adopts the zap logger open sourced by Uber. As one the most powerful log components in the Go language ecosystem, zap incorporates an http_handler.go module so that you can view the current log level and dynamically change the log level via an HTTP interface.
Milvus listens the HTTP service provided by the
9091 port. Therefore, you can access the
9091 port to take advantage such features as performance debugging, metrics, health checks. Similarly, the
9091 port is reused to enable dynamic log level modification and a
/log/level path is also added to the port. See the log interface PR or the source code for more information.
This section provides instructions on how to dynamically change log levels without the need the restarting the running Milvus service.
Ensure that you can access the
9091 port of Milvus components.
Suppose the IP address of the Milvus proxy is
You can first run
$ curl -X GET 192.168.48.12:9091/log/level to check the current log level of the proxy.
Then you can make adjustments by specifying the log level. Log level options include:
The following example code changes the log level from the default log level from
$ curl -X PUT 192.168.48.12:9091/log/level -d level=error
Like the article? Spread the word
An anatomy of the mechanism behind tunable consistency levels in the Milvus vector database.
Learn about user authentication and encryption in transit in Milvus.
Learn about the four levels of consistency - strong, bounded staleness, session, and eventual supported in the Milvus vector database.